Combustion and Flame 136 (2004) 383–389

www.elsevier.com/locate/jnlabr/cnf
Brief Communication

Bifurcation of flame structure in a lean-premixed

swirl-stabilized combustor: transition from stable to
unstable flame
Ying Huang and Vigor Yang ∗
Department of Mechanical Engineering, Pennsylvania State University, University Park, PA 16802, USA
Received 25 April 2003; received in revised form 20 October 2003; accepted 5 November 2003

Abstract
The present work addresses unsteady flame dynamics in a lean-premixed swirl-stabilized combustor, with at-
tention focused on the transition of flame structure from a stable to an unstable state. It was found that the inlet
temperature and equivalence ratio are the two most important variables determining the stability characteristics of
the combustor. A slight increase in the inlet mixture temperature across the stability boundary leads to a sudden
increase in acoustic flow oscillation. One major factor contributing to this phenomenon is that as the inlet mixture
temperature increases, the flame, which is originally anchored in the center recirculation zone, penetrates into the
corner recirculation zone and flashes back, due to the increased flame speed. As a consequence, the flame is stabi-
lized by both the corner- and the center-recirculating flows and exhibits a compact enveloped configuration. The
flame flaps dynamically and drives flow oscillations through its influence on unsteady heat release. This problem
has not previously been studied mechanistically. The results improve our understanding of the mechanisms of
initiation and sustenance of combustion instabilities in gas-turbine engines with lean-premixed combustion.
 2003 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

1. Introduction an unstable operation (characterized by a limit cycle

with large oscillation) is observed. The self-excited
It is well established that the flow and flame dy- large unsteady flow oscillations in combustors, which
namics in a combustion chamber can change dra- are usually referred to as combustion instability, have
matically as the governing parameters pass through hindered the development of gas-turbine engines with
their critical values at which bifurcation points are lo- lean-premixed (LPM) combustion for many years.
cated. Combustion processes alone may or may not Understanding of the mechanisms responsible for in-
exhibit bifurcation phenomena, but when they take ducing bifurcation is important for passive and active
place in the presence of the nonlinear behavior of control of combustion instability [3,4].
the chamber dynamics, this sort of characteristic is Several experimental studies [1–8] have been con-
indeed observed in many combustion devices [1–3]. ducted to investigate combustion dynamics with bi-
Sometimes, when bifurcation takes place, which may furcation phenomena. Culick and colleagues [1,4] in-
arise from disturbances of the governing parameters, vestigated the hysteresis behavior of combustion in-
transition from a stable operation (characterized by a stability in a dump combustor as a function of the
limit cycle with small oscillation or no oscillation) to mixture equivalence ratio, in which several attractors
coexist for a given parameter value and the transi-
tion from a stable to an unstable state and its reverse
* Corresponding author. occur at different critical parameter values. Lieuwen
E-mail address: [email protected] (V. Yang). [3] studied the limit-cycle oscillations in a gas-turbine

0010-2180/$ – see front matter  2003 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
doi:10.1016/j.combustflame.2003.10.006
384 Y. Huang, V. Yang / Combustion and Flame 136 (2004) 383–389

Fig. 1. Schematic of a model swirl-stabilized gas-turbine

combustor (after Broda et al. [2]; used by permission of the
publisher).

combustor. It was suggested that cyclic variability is

caused by background noise. The inlet velocity not
only plays an important role in determining the sta-
bility characteristics of the combustor, but also af-
fects the amplitudes of the oscillations. Broda et al.
[2] performed an experimental study of combustion
dynamics in a swirl-stabilized gas-turbine combustor.
The system consists of a single-swirl injector, an ax-
isymmetric chamber, and a choked nozzle, as shown
schematically in Fig. 1. The estimated swirl number
is 0.76 for a 45◦ swirler with a constant chord (c)
and angle (φ). Natural gas is injected radially from
the center body through 10 holes immediately down-
stream of the swirler vanes. The fuel/air mixture is
assumed to be well mixed before entering the com-
bustor. A broad range of equivalence ratio and inlet
air temperature was considered systematically. Fig-
ure 2 shows stability maps as a function of inlet air Fig. 2. Stability maps as a function of inlet air temperature
temperature and equivalence ratio. Instabilities occur and equivalence ratio (after Seo [8]; used by permission of
only when the inlet air temperature is greater than a the publisher).
threshold value Tin∗ around 660 K and the equivalence
ratio falls into the range between 0.5 and 0.7. Figure 3 imuthal structures. Angelberger et al. [11] conducted
shows typical photographic images of a stable and an a two-dimensional simulation of a premixed dump
unstable flame with an equivalence ratio 0.6. As the combustor with acoustic forcing. Fureby [12] inves-
inlet temperature increases and exceeds the threshold tigated the combustion instabilities of a dump com-
value Tin∗ , the flame structure transforms from a sta- bustor and a model afterburner. Vortex shedding was
ble to an unstable state, and the amplitude of pressure found to be a main attribute of the combustion insta-
oscillation increases and reaches another limit cycle. bilities observed. For LPM combustion with swirling
This kind of bifurcation in flame structure, as the inlet flows, a comprehensive numerical study was recently
temperature varies, will be investigated numerically performed by the authors [13], demonstrating the ca-
in the present study. pability of LES in treating detailed combustion dy-
Since the flowfield of concern is highly unsteady namics.
and dominated by turbulent motions that need to be The purpose of this work is to study the effects of
adequately resolved computationally, a large eddy inlet flow temperature on the flame bifurcation phe-
simulation (LES) technique is adopted. Several at- nomenon, simulating the experimental conditions re-
tempts have been made to study combustion dy- ported in Refs. [2] and [8]. In an earlier paper [13],
namics using LES. Thibaut and Candel [9] stud- the flame dynamics under an unstable condition was
ied the flashback phenomenon in a backward-facing studied. The present analysis will focus on the transi-
step configuration using two-dimensional LES. The tion from a stable to an unstable flame when the inlet
mechanism of flashback associated with combustion temperature exceeds a threshold value. The situation
dynamics was investigated. Kim et al. [10] investi- with a stable flame will also be addressed to serve as
gated a swirl-stabilized gas-turbine combustor flow. a reference. The basis of the analysis is the LES code
Their work suggests that high swirl results in a very previously developed for investigating LPM combus-
complex vortex-shedding pattern with significant az- tion instability [13]. Various fundamental processes
 Y. Huang, V. Yang / Combustion and Flame 136 (2004) 383–389 385

ergy. The subgrid-scale terms are modeled using a

compressible-flow version of the Smagorinsky model
suggested by Erlebacher et al. [14]. The damping
function of Van-Driest is used to take into account
the inhomogeneities near the wall. For treating turbu-
lent combustion within the context of LES, combus-
tion models are often needed on the subgrid scales.
A level-set flamelet library approach, which has been
successfully applied to study premixed turbulent com-
bustion [13,15], is used here.
Boundary conditions must be specified to com-
plete the formulation. The no-slip and adiabatic con-
ditions are enforced along all of the solid walls. At
the inlet boundary, the mass flow rate and temper-
ature are specified. The pressure is obtained from a
one-dimensional approximation to the axial momen-
tum equation, i.e., ∂p/∂x = −ρ∂u/∂t − ρu∂u/∂x.
The mean axial-velocity distribution follows the one-
seventh power law by assuming a fully developed tur-
bulent pipe flow. The radial and azimuthal velocities
are determined from the swirler vane angle. Turbu-
lence properties at the inlet are specified by super-
imposing broadband disturbances with an intensity of
15% of the mean quantity onto the mean velocity pro-
files. The nonreflecting boundary conditions proposed
by Poinsot and Lele [16] are applied at the exit bound-
ary.
The resultant governing equations and boundary
conditions are solved numerically by means of a
density-based, finite-volume methodology. The spa-
tial discretization employs a second-order, central-
differencing method in generalized coordinates.
A fourth-order matrix dissipation with a total-varia-
tion-diminishing switch developed by Swanson and
Turkel [17] and tested by Oefelein and Yang [18]
is included to ensure computational stability and to
prevent numerical oscillations in regions with steep
gradients. Temporal discretization is obtained using a
four-step Runge–Kutta integration scheme. A multi-
block domain decomposition technique along with
static load balance is used to facilitate the implemen-
tation of parallel computation with message passing
interface at the domain boundaries. The theoretical
Fig. 3. Top: Photographic images of stable and unstable
and numerical framework described above has been
flames. Bottom: Pressure–time trace (after Seo [8]; used by
permission of the publisher). validated by Huang et al. [13] and Apte and Yang
[19] against a wide variety of flow problems in order
responsible for the flame transition from a stable to an to establish its creditability and accuracy.
unstable state, such as high-temperature mixture fill-
ing, flame trapping, and the vortex flashback process,
are carefully identified and quantified. 3. Results and discussion

The model combustor shown in Fig. 1 and de-

2. Theoretical formulation and numerical method scribed in the preceding section is considered in the
present study. More detailed information about the
The formulation is based on the Favre-filtered experimental facility and results can be found in
conservation equations of mass, momentum, and en- Ref. [2]. The chamber measures 45 mm in diameter
386 Y. Huang, V. Yang / Combustion and Flame 136 (2004) 383–389

and 235 mm in length. The baseline condition in- respect to the acoustic velocity at the interface be-
cludes an equivalence ratio of 0.573 and a chamber tween the inlet and combustor. The entire process is
pressure of 0.463 MPa. The mass flow rates of natural dictated by the temporal evolution and spatial dis-
gas and air are 1.71 and 50.70 g/s, respectively. The tribution of the flame front, which moves back and
inlet flow velocity is 86.6 m/s and the corresponding forth under the influences of the vortical motion (indi-
Reynolds number based on the inlet flow velocity and cated by the concentrated streamlines) in the chamber.
height of the inlet annulus is 35,000. A new vortex begins to shed from the center body at
Because of the enormous computational effort re- θ = 90◦ , accompanying a higher local flow velocity.
quired for calculating the flowfield in the entire cham- As the vortex moves downstream (θ = 180◦ –270◦ ),
ber, only a cylindrical sector with periodic bound- it distorts the flame front or even produces a sep-
ary conditions specified in the azimuthal direction arated flame pocket. At the same time, the higher
is treated herein. The analysis, despite the lack of speed mixture pushes the flame downstream. When
vortex-stretching mechanism, has been shown to be
able to capture the salient features of the turbulent
flowfields and unsteady flame propagation [9,13]. The
computational domain includes the upstream half of
the chamber and part of the inlet duct. The entire
grid system consists of 375 × 140 points along the
axial and radial directions, respectively, of which
75 axial points are used to cover the inlet section.
The largest grid size falls in the inertial subrange Fig. 4. Mean temperature contours and streamlines of stable
of the turbulent energy spectrum, based on the in- flame.
let Reynolds number. The grids are clustered in the
shear-layer regions downstream of the dump plane
and near the solid walls to resolve the shear-layer and
near-wall gradients. The computational domain is di-
vided into 17 blocks and the analysis was conducted
on a distributed-memory parallel computer with each
block calculated on a single processor.
Stable flame evolution was first obtained for an in-
let mixture temperature of 600 K (below the threshold
value Tin∗ for the onset of combustion oscillation). The
flame bifurcation phenomenon was then investigated
by increasing the inlet temperature from 600 to 660 K.
The mean chamber pressure is 0.463 MPa. Figure 4
shows the mean temperature contours and pseudo-
streamlines on the x–r plane based on the mean axial
and radial velocity components for a stable flame.
A central torodial recirculation zone (CTRZ) is es-
tablished in the wake of the center body under the
effects of the swirling flow. The CTRZ, a form of vor-
tex breakdown, serves as a flame stabilization region,
where hot products are mixed with the incoming mix-
ture of air and fuel. In addition, as a result of the sud-
den increase in combustor area, a corner recirculation
zone is formed downstream of the backward-facing
step.
The calculated pressure and velocity fields ex-
hibit small-amplitude fluctuations with a dominant
harmonic mode at 3214 Hz, corresponding to the fre-
quency of the vortex shedding from the center body.
Figure 5 presents the flame evolution and vortex shed-
ding process in the upstream region of the chamber
over one cycle of oscillation. The pressure and ve-
locity are measured at the middle point of the inlet Fig. 5. Stable flame evolution over one cycle of oscillation
annulus exit. The phase angle θ is referenced with (3214 Hz): temperature contours and streamlines.
 Y. Huang, V. Yang / Combustion and Flame 136 (2004) 383–389 387

the vortex moves away from the flame (θ = 360◦ ) and

dissipates into small-scale structures, the flame front
propagates upstream (since the higher speed mixture
is convected downstream) and interacts with another
incoming vortex. During this process, a new vortex
appears at the corner of the center body and another
cycle repeats.
The inlet temperature has enormous effects on the
flame dynamics in the system. On the one hand, when
the inlet temperature increases, for a fixed mass flow
rate, the flow velocity also increases and pushes the
flame downstream. On the other hand, the increased
inlet temperature leads to an increase in the flame
speed and consequently causes the flame to propagate
upstream. In addition, flashback may occur near the
wall due to the small local flow velocity. The com-
bined effects of flow acceleration, flame-speed en-
hancement, and flashback determine the final form of
the flame structure.
In the present study, as the inlet temperature in-
creases from 600 to 660 K, flame bifurcation takes
place. The flame originally anchored in the center
recirculation zone penetrates into the corner recir-
culation zone and flashes back. Consequently, the
flame is stabilized by both the corner- and the center-
recirculating flows and forms a compact enveloped
configuration. The flame flaps dynamically and drives
flow oscillations through its influence on unsteady
heat release. At the same time, the pressure oscilla-
tion increases and reaches another limit cycle with a
much larger amplitude. The entire bifurcation process
can be divided into three stages: high-temperature
mixture-filling process, flame-trapping process, and
vortex-flashback process, as shown in Fig. 6, in which
t = 0 ms denotes the time at which the inlet mixture
temperature starts to increase from 600 to 660 K.
Figures 6a–6c show the high-temperature mixture-
filling process. As the inlet mixture temperature in-
creases, the flow speed increases due to the decreased
density for a fixed mass flow rate. As a result, the orig-
inal low-temperature mixture is pushed downstream
toward the flame. Although a flashback phenomenon Fig. 6. Transition from stable to unstable flame with in-
is observed near the wall, the high-temperature mix- creased inlet temperature from 600 to 660 K.
ture has not reached the flame front near the wall and
the flame speed remains unchanged at this stage.
Figures 6d and 6e show the flame-trapping process. the backward-facing step approaches the flame front
Once the high-temperature mixture reaches the flame in the corner recirculation zone and then pushes it
front, with the help of the increased flame speed, the toward the dump plane. At the same time, a small
near-wall flashback overshadows the flow accelera- flame pocket is produced and separated from the
tion effects. Consequently, the flame front penetrates main stream. After this vortex is convected down-
into the corner recirculation zone and is trapped by stream and passes through the flame, another vortex
the local vortical motion. approaches and interacts with the flame. This process
In the vortex-flashback process, as shown in continues and eventually the fresh reactants in the
Figs. 6f–6h, the flame propagates upstream under corner recirculation zone are completely burned. The
the influence of the vortical motion. A counterclock- flame is stabilized by both the corner- and the center-
wise rotating vortex originally shed from the edge of recirculating flows and its overall length is substan-
388 Y. Huang, V. Yang / Combustion and Flame 136 (2004) 383–389

tially reduced. This situation renders the combustor the quenching distance. This criterion, however, is
more prone to instabilities according to the Rayleigh qualitatively correct only for isothermal walls and is
criterion [13,20], since considerable heat is released not applicable for adiabatic walls due to the lack of a
within a short distance close to the chamber head-end quenching distance. Another criterion, valid for both
(i.e., the acoustic anti-node point). adiabatic and isothermal walls, was recently proposed
In our previous work [13], a fully three-dimension- by Kurdyumov et al. [22]. Flashback occurs if the
al LES study was performed to investigate the dy- Karlovitz number, defined as αA/SL 2 , with α being
namics of an unstable flame. Several mechanisms the thermal diffusivity and A the velocity gradient at
responsible for driving combustion instabilities in the the wall, is less than a critical value. Although this
chamber were identified and quantified. The energy criterion is formulated for laminar flows, the result
cascade from chemical reactions in the flame zone can be qualitatively extended to flames in turbulent
to acoustic fluctuations in the chamber was found to boundary layers. In the present case, the flame speed
be characterized by a feedback closed-loop process increases as the inlet temperature increases. Conse-
which includes the mutual coupling between acoustic quently, the flame is more prone to flashback through
motion, vortex shedding, flame propagation, and heat- the wall boundary layers according to Kurdyumov’s
release oscillation. criterion. Flashback arising from local flow reversal
Once the flame becomes unstable when the inlet has also been investigated by many researchers (see,
flow temperature exceeds the critical value Tin∗ , it be- for example, Refs. [9] and [23]). Large vortical struc-
comes rather difficult to reestablish stable operation tures and turbulent flame speed play important roles
unless the inlet temperature is reduced to a level sig- in this kind of phenomenon. The latter is essential be-
nificantly lower than Tin∗ . This phenomenon is com- cause it controls the rate of mixture consumption.
monly referred to as hysteresis and has been experi- For lean-premixed combustion, the laminar flame
mentally observed by many researchers (see, for ex- speed SL increases with an increase in the equiva-
ample, Ref. [2]). The occurrence of hysteresis under lence ratio φ. Thus, increases in the equivalence ra-
the current circumstance may be explained as follows. tio and inlet temperature exert similar effects on the
During unstable combustion, the corner recirculation flame evolution. However, the chemical reaction rate
zone is filled with high-temperature products and the and heat release are much more sensitive to variations
chamber wall in this region is heated to reach the lo- in the equivalence ratio under lean conditions than
cal flame temperature. To recover stable operation, under stoichiometric conditions. Moreover, near the
the cold flow needs not only to extinguish the flame lean blowout limit, perturbations in the equivalence
stabilized by the corner-recirculating flow through en- ratio φ can cause periodic extinction of the flame. As
trainment or flame lift-off, but also to offset the ef- a result, the equivalence ratio oscillation under lean
fects of high-temperature wall, which tend to increase conditions is prone to inducing flow oscillation [5]
the local gas temperature and inhibit extinction and and subsequently increases turbulent velocity fluctu-
near-wall flashback. Consequently, a much lower in- ation ν  . This suggests that a lean-premixed turbulent
let temperature is required to regain stable operation. flame is more susceptible to flashback, since the tur-
Numerical simulation of the hysteresis phenomenon bulent flame speed ST increases not only with the
necessitates a refined treatment of flame extinction laminar flame speed SL but also with turbulent veloc-
and wall boundary conditions, a subject for subse- ity fluctuation ν  [24]. The result helps explain why
quent research. the transition from a stable to an unstable state as de-
In light of the above observations, we conclude scribed in Ref. [2] occurs only when the equivalence
that the flashback phenomenon dictates the flame bi- ratio falls in the range between 0.5 and 0.7.
furcation process. Flashback in premixed combustion Since the flame bifurcation is largely determined
has been the subject of a number of experimental, by the flashback phenomenon in the corner recircu-
analytical, and numerical studies in the past. Its oc- lation zone in the present case, one effective way to
currence is usually attributed to two mechanisms. The avoid its occurrence is to inject cold flow into that
first involves flame propagation in the boundary layer region. This procedure suppresses the local flame up-
along a solid wall, where the local velocity diminishes stream propagation and consequently leads to a much
toward the surface. The second mechanism is associ- more stable system.
ated with flow reversal, which is usually caused by
vortical motions or acoustic oscillations. Both mech-
anisms are observed in the present case. 4. Conclusions
A criterion for the occurrence of near-wall flash-
back was proposed by Lewis and Von Elbe [21], who The unsteady flame dynamics in a lean-premixed,
state that flashback occurs if the velocity gradient at swirl-stabilized combustor has been studied using a
the wall is less than the ratio of the flame speed and large-eddy-simulation technique along with the level-
 Y. Huang, V. Yang / Combustion and Flame 136 (2004) 383–389 389

set flamelet library approach. Special attention was [3] T. Lieuwen, J. Propuls. Power 18 (2002) 61–67.
focused on the transition of flame structure from a [4] G. Isella, C. Seywert, F.E.C. Culick, E.E. Zukoski,
stable to an unstable state. Results indicate that the Combust. Sci. Technol. 126 (1997) 381–388.
inlet temperature and equivalence ratio are the key [5] T. Lieuwen, H. Torres, C. Johnson, B.T. Zinn, J. Eng.
Gas Turb. Power 123 (2001) 182–189.
parameters determining the stability characteristics of
[6] C.E. Johnson, Y. Neumeier, T.C. Lieuwen, B.T. Zinn,
the combustor. A slight increase in the inlet airflow
Proc. Combust. Inst. 28 (2000) 757–763.
temperature across the stability boundary leads to a [7] J.G. Lee, K. Kim, D.A. Santavicca, Proc. Combust.
sudden increase in the chamber flow oscillations. One Inst. 28 (2000) 739–746.
major factor contributing to this phenomenon is that [8] S. Seo, PhD thesis, Department of Mechanical Engi-
as the inlet mixture temperature increases, the flame neering, The Pennsylvania State University, University
originally anchored in the center recirculation zone Park, PA, 1999.
flashes back through the wall boundary layers and the [9] D. Thibaut, S. Candel, Combust. Flame 113 (1998) 53–
vortical flow downstream of the dump plane. As a re- 65.
sult, the flame becomes stabilized by both the corner- [10] W.W. Kim, S. Menon, H.C. Mongia, Combust. Sci.
Technol. 143 (1999) 25–62.
and the center-recirculating flows. The flame then
[11] C. Angelberger, D. Veynante, F. Egolfopoulos, Flow
flaps dynamically and drives flow oscillations through Turb. Combust. 65 (2000) 205–222.
its influence on unsteady heat release. Much infor- [12] C. Fureby, Proc. Combust. Inst. 28 (2000) 783–791.
mation has been obtained to shed light on the funda- [13] Y. Huang, H.G. Sung, S.Y. Hsieh, V. Yang, J. Propuls.
mental mechanisms of the initiation and sustenance of Power 19 (2003) 782–794.
combustion instabilities in gas-turbine engines with [14] G. Erlebacher, M.Y. Hussaini, C.G. Speziale, T.A.
lean-premixed combustion. The present study, how- Zang, J. Fluid Mech. 238 (1992) 155–158.
ever, did not consider the influence of stretch rate [15] L.D. de Lageneste, H. Pitsch, Annual Research Briefs,
on flame dynamics, which may alter the local flame Center for Turbulence Research, Stanford University,
Stanford, CA, 2000.
structure and cause flame extinction or lift-off. This
[16] T. Poinsot, S. Lele, J. Comput. Phys. 101 (1992) 104–
will be explored in subsequent work. 129.
[17] R.C. Swanson, E. Turkel, J. Comput. Phys. 101 (1992)
292–306.
Acknowledgment [18] J.C. Oefelein, V. Yang, J. Propuls. Power 14 (1998)
843–857.
The work reported in this paper was sponsored [19] S. Apte, V. Yang, J. Fluid Mech. 477 (2003) 215–225.
by the Air Force Office of Scientific Research, Grant [20] J.W.S. Rayleigh, The Theory of Sound, vol. II, Dover,
New York, 1945, reprinted.
F49620-99-1-0290.
[21] B. Lewis, G. von Elbe, Combustion, Flames and Explo-
sions of Gases, third ed., Academic Press, New York,
1987, pp. 233–236.
References [22] V.N. Kurdyumov, E. Fernandez, A. Linan, Proc. Com-
bust. Inst. 28 (2000) 1883–1889.
[1] P. Knoop, F.E.C. Culick, E.E. Zukoski, Combust. Sci. [23] H.N. Najm, A.F. Ghoniem, J. Propuls. Power 10 (1994)
Technol. 123 (1997) 363–376. 769–776.
[2] J.C. Broda, S. Seo, R.J. Santoro, G. Shirhattikar, [24] A.N. Lipatnikov, J. Chomiak, Prog. Energy Combust.
V. Yang, Proc. Combust. Inst. 27 (1998) 1849–1856. Sci. 28 (2002) 1–74.